[openstack-dev] [nova] [infra] The same SRIOV / NFV CI failures missed a regression, why?
fungi at yuggoth.org
Tue Apr 12 17:57:21 UTC 2016
On 2016-04-05 00:45:20 -0400 (-0400), Jay Pipes wrote:
> The proposal is to have the hardware companies donate hardware and
> sysadmins to setup and maintain a *single* third-party CI lab
> environment running the *upstream infra CI toolset* in one
> datacenter at first, moving to multiple datacenters eventually.
> This lab environment would contain hardware that the vendors
> intend to ensure is functionally tested in certain projects --
> mostly Nova and Neutron around specialized PCI devices and SR-IOV
> NICs that have zero chance of being tested functionally in the
> cloudy gate CI environments.
This is great. I always love to see increased testing of OpenStack
(and often more insights come from setting up the test environment
and designing the tests than result from running them on proposed
> The thing I am proposing the upstream Infra team members would be
> responsible for is guiding/advising on the creation and
> installation of the CI tools and helping to initially get the CI
> system reporting to the upstream Jenkins/Zuul system. That's it.
> No long-term maintenance, no long-term administration of the
> hardware in this lab environment. Just advice and setup help.
We already have a vibrant and active community around these tools
and concepts and I welcome any additional participation there.
> The vendors would continue to be responsible for keeping the CI
> jobs healthy and the lab environment up and running. It's just
> instead of 12 different external CI systems, there would be 1
> spawning jobs on lots of different types of hardware. I'm hoping
> that reducing the number of external CI systems will enable the
> vendors to jointly improve the quality of the tests because they
> will be able to focus on creating tests instead of keeping 12
> different CI systems up and running.
> Hope that better explains the proposal.
It does. I'm glad to see the implications in the original proposal
of being partially staffed by the foundation turned out not to be
integral. Your call for a "single CI system" also seemed to imply
combining this and the upstream CI rather than simply combining
multiple third-party CI systems into a single third-party CI system,
which I now get is not actually the case. Thanks for clarifying!
More information about the OpenStack-dev